Download A Note on the Comparative Statics of Optimal Procurement Auctions

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Internal Revenue Code section 1031 wikipedia , lookup

Capital gains tax in Australia wikipedia , lookup

Short (finance) wikipedia , lookup

Transcript
A Note on the Comparative Statics of Optimal
Procurement Auctions∗
Gonzalo Cisternas†
Nicolás Figueroa‡
January 2008
Abstract
We consider a buyer who must procure a service from one of n potential sellers, whose production
costs are private information. We find a necessary and sufficient condition such that a distributional
upgrade on a seller’s cost distribution implies a lower expected procurement cost for the buyer. We
also show that even under the strongest assumption about this upgrade made in the literature so
far, the seller can be worse off, even if this upgrade is costless. Keywords: Procurement Auctions,
Mechanism Design, Distributional Upgrade. JEL D44, C7, C72.
1. Introduction
In many procurement circumstances, it is possible for one of the sellers to improve his cost distribution
before the procurement mechanism actually takes place. This change could be exogenous, for example due
to a change in its input prices, or endogenous, due to a costly investment that improves the production
technology. It is very important to understand the change in the buyer’s expected cost and the sellers
expected profit when such a distributional improvement occurs. If the seller benefits very little from it,
then he will not invest much in researching a new technology, and if he is worse-off after a change, he will
likely block opportunities for an exogenous improvement. In the same spirit, if the buyer benefits the
most, then it is likely she will subsidize the investment. A situation where a seller can improve his cost
distribution therefore sheds light on the the motivation of a downstream firm to buy one of his upstream
providers. If such a provider does not have the incentives to improve his technology and participate in
the subsequent procurement auction, but the buyer would benefit greatly by him doing so, then the buyer
will likely try to buy the provider and do the investment himself. To start an analysis of these important
economic questions, it is crucial to study the changes in the sellers cost and the buyers profit when such
an improvement occurs. We do this in a simple model of static procurement.
∗ We are specially grateful to Felipe Balmaceda for his constructive comments and suggestions. This research was partially
supported by the Complex Engineering Systems Institute and by Fondecyt Grant No 1107059.
† Centro de Economı́a Aplicada, DII, Universidad de Chile, República 701, Santiago, Chile; [email protected]
‡ Centro de Economı́a Aplicada, DII, Universidad de Chile, República 701, Santiago, Chile; [email protected]
1
We consider a buyer who must procure a service from one of n potential sellers, whose production costs
are private information. A simple reformulation of the model in [6] allows to characterize the optimal
mechanism used by the buyer, the expected procurement cost and the expected profits of a seller.
For the buyer, facing a better seller is good since there is a higher probability of her having low costs
but, on the other hand, it may be bad since having a better distribution can imply that the informational
rent she can extract is also higher.
For the seller, having a better distribution is good since, ceteris paribus, it increases her probabilities
of winning the auction and the informational rent she can extract. However, since this better distribution
is observed by the buyer and the mechanism is changed against the better seller, there is a negative effect
associated to it.
For the reasons outlined above, the comparative statics for both the buyer and the seller with respect
to the cost distribution of a seller are not obvious. We are interested in the circumstances under which it
is desirable for the buyer to face one “better” seller (with a better cost distribution) and for the seller to
have such a cost distribution. In other words, we are interested in the comparative statics of the buyer’s
expected cost and a seller’s expected profit with respect to a distributional upgrade on a seller.
Up to now, this problem has not been solved. That is, the right notion of distributional improvement
that guarantees that the buyer is better off has not been characterized. We provide a natural and
weak necessary condition on the distributional upgrade under which the buyer is better off. On the other
hand, with a very simple counterexample, we show that for even for the strongest concept of distributional
improvement used in the literature, the seller can be worse off when her cost distribution improves, even
if this improvement is costless. The buyer biases the mechanism against the better seller so much, that
the potential gains of a better cost distribution and a higher informational rent are offset by a decreased
chance of winning due to this adjusted mechanism.
This last result strengthens the results of underinvestment in Dasgupta [3] in a symmetric environment,
and also the result in Arozamena and Cantillon [1], where underinvestment is the result of a strong reaction
by other bidders in a first price auction. Here, the underinvestment result is extreme, since the socially
efficient investment for a costless technology is infinity and the seller would choose to invest 0. Moreover,
the seller would be strictly worse off in case of an exogenous improvement in her distribution.
2. Model
Consider a buyer who wants to procure a good or service and faces n potential suppliers indexed by
i = 1, ..., n. If the buyer decides to carry out the task by himself, it would cost him an amount of money
c0 ≥ c. Suppliers’ costs to perform the task (which are private information) are distributed independently
across firms. Firm i obtains her cost from a differentiable distribution Fi (·), i ≥ 2, with support C ≡ [c, c̄].
However, competitor 1 (from now on the upgrader ) draws his cost from a differentiable distribution F (·, I)
with the same support as before. I is a parameter that indexes the supplier’s efficiency, and we assume
that, as I ≥ 0 increases, the distribution improves. For notational convenience we use fi (·) ≡ Fi0 (·) if
i ≥ 2, and ∂F
∂c (c, I) for the upgrader.
We make the standard regularity assumption (first stated in [6]), which guarantees that the optimal
mechanism can be found using pointwise maximization.
2
Assumption 1 For every i ≥ 2 and I ≥ 0, the functions Ji (c) = c +
increasing.
Fi (c)
fi (c)
and JI (c) = c +
F (c,I)
(c,I)
∂F
∂c
are
For technical reasons, we also need:
Assumption 2 For every c ∈ C, I 7→ JI−1 (c) is differentiable.
There are several “distributional improvements” that may apply to the context presented here. We
introduce two widely-used notions, the first one being the most commonly used in statistics and economics:
Definition 3 (First Order Stochastic Dominance): We will say that {F (·, I)}I∈IR+ is a family of
distributional improvements in the sense of first order stochastic dominance (FOSD) if, for every fixed
c ∈ C, F (c, ·) is increasing. In other words, the probability of obtaining a cost below c ∈ C is increasing
in I.
The next one has been of great use in the auction literature (see for example [4]):
Definition 4 (Monotone Likelihood Ratio Property): We will say that {F (·, I)}I∈IR+ is a family of
distributional improvements in the sense of the monotone likelihood ratio property (MLRP) if, for every
I 0 < I ∈ IR+ and c0 < c ∈ C,
∂F 0 0
∂c (c , I )
∂F
0
∂c (c, I )
<
∂F 0
∂c (c , I)
∂F
∂c (c, I)
(1)
That is, as I increases, it is more likely to obtain lower costs relative to higher ones. This condition is
equivalent to (c, I) 7→ ∂F
∂c (c, I) being log-submodular.
The following well-known result relates both definitions and shows that MLRP is stronger than FOSD:
Lemma 5 If {F (·, I)}I∈IR+ is a family of distributional improvements in the sense of MLRP, then, it is
a family of distributional improvements in the sense of FOSD.
Proof. See, for example, [4].
F (c,I)
Observation: Another well-known result shows that MLRP also implies that ∂F
is increasing in I
∂c (c,I)
for all c ∈ C. This term corresponds to the informational rent of a seller of type c, and the fact that it
increases with I makes the effect of a better seller unclear for the buyer: a better seller has lower costs
but also extracts a higher informational rent.
Finally, define C n = {cn = (c1 , ..., cn )| ci ∈ C ∀i = 1, ..., n} and assume that for i ≥ 2, fi (·) > 0 and
∀I ≥ 0, ∂F
∂c (·, I) > 0, a.e. in C.
3
3. Comparative Statics
We now consider an upgrader with cost distribution F (·, I), and perform comparative statics over the
procurement cost with respect to the parameter I. The buyer’s problem is to choose transfer functions
ti : C n → IR (payments to the sellers) and winning probability functions qi : C n → [0, 1] (probabilities of
buying), i = 1, ..., n. Under the regularity assumptions, it is direct that the expected optimal mechanism
corresponds to (see [6])
qi∗ (c1 , ..., cn ) =
1 JI (c1 ) ≤ min{c0 , Ji (ci )| i ≥ 2}
0
∼
(2)
1 Ji (ci ) < min{c0 , JI (c1 ), Jl (cl )| l 6= i, l ≥ 2}
0
∼
(3)
q1∗ (c1 , ..., cn ) =
i = 2, ..., n.
which yields a procurement cost of:

Z
C(I)
=
Cn

JI (c1 )q1∗ (cn ) +
X
Jl (cl )ql∗ (cn ) + c0 1 −
i≥1
l≥2


Y
∂F
(c1 , I) 
fj (cj ) dcn
qi∗ (cn )
∂c1

X
j≥2
(4)
On the one hand, an increase in I implies that the distribution F (·, I) puts more weight on low-cost
realizations, therefore reducing the total cost C(I). On the other hand, due to informational asymmetries,
the buyer pays (in expected terms) an amount JI (c) to the upgrader. This term can be increasing in I for
a sufficiently strong distributional upgrade (for example one that satisfies MLRP), therefore increasing
the total cost C(I). Then, for a fixed mechanism, an increase in I could affect C(I) in both ways.
However, under these new circumstances, the buyer adapts the mechanism (through a change in the
optimal rules qi∗ ) and can give a disadvantage to a “better” seller who has a bigger virtual cost, tilting
the balance and decreasing the total cost when I increases.
Our main purpose is to establish which conditions on the family {F (·, I)}I≥0 imply that the expected procurement cost is reduced. The main proposition, stated below, shows that a natural and weak
condition like FOSD suffices to imply the result.
Proposition 6 Suppose that for every c ∈ C the function F (c, ·) is differentiable. A sufficient pointwise
conditions on the the family {F (·, I)}I≥0 under which the expected procurement cost decreases is:
∀I ≥ 0, ∀c ∈ [c, JI−1 (c0 )],
∂F
(c, I) ≥ 0
∂I
(5)
As a consequence, if the mentioned family satisfies FOSD, the expected procurement cost decreases when
facing a better competitor.
4
Proof. See Appendix.
Observation: Condition (5) can indeed be much weaker than FOSD, since it must only be satisfied
in the interval [c, JI−1 (c0 )], which corresponds to the set where the upgrader has a “chance” of winning,
since JI (c) < c0 in this region. This interval is small if c0 , the reservation cost, is small, since in that
case the purchase does not occur often because of the seller has an attractive opportunity of doing the
project himself.
If a property like MLRP is satisfied, then JI (c) is increasing in I for all c. This implies that the
interval [c, JI−1 (c0 )] shrinks with I, making the condition weaker as I increases. This is true since as I
increases, the seller with a better cost distribution faces more disadvantageous mechanisms, and therefore
the region where he has a chance of being assigned the project is reduced.
As we can see, the tradeoff mentioned in the introduction (a buyer likes a better seller since he has
in average lower costs, but the other hand dislikes one, since he can extract higher informational rents)
always works in the buyer’s favor. The total surplus generated is clearly bigger with a better seller, and
the possibly higher informational rents he is able to obtain are not enough to offset this fact. This is
true since the buyer modifies the mechanism, giving an ex-ante disadvantage to the better seller, and
therefore controlling the rent he is able to extract.
We conclude by pointing out that the strength of this last effect, where the buyer biases the mechanism
against the better seller to extract more rent, can be enough to make the seller worse off, even if the
distributional upgrade is for free. This last fact is particularly important since it sheds light on the
make-or-buy problem. The buyer would like the seller to invest in new technologies that improve his
cost distribution, but he would have to give incentives (either monetary or through the commitment
to a less disadvantageous mechanism) for this to happen. Moreover, if investment is non-observable or
commitment not possible, the buyer will have incentives to buy the provider, invest and produce the good
or service himself.
1
Example:Suppose n = 2, C = [0, 1] and c0 = +∞. Consider F2 (c) = c and F (c, I) = c 1+I , I ≥ 0.
This family of distributions corresponds to a seller which upgrades the reliability of his technology, that
is the maximum of I + 1 independent draws follows a uniform distribution in [0,1]. Moreover, this family
satisfies MLRP and, as a consequence, FOSD. The upgrader’s expected utility when his distribution is
F (·, I) corresponds to1
Z
Z
∂F
(c, I)dc = Q∗ (c)F (c, I)dc
Π(I) = Π(c, c)
∂c
C
∗
with Q (c) =
R
∗
C
∗
q (c, s)f (s)ds. Using that q (c, s) = 1 ⇔ JI (c) ≤ J2 (s) (from (2) and (3)), JI (c) = c(2+I)
C
and J2 (c) = 2c, thus J2−1 (JI (c)) =
c(2+I)
2
and we have
2
2+I
Z2+I 1+I
1
2
c(2 + I) 2+I
(1 + I)2
Π(I) =
c dc =
1−
2
(2 + I)(3 + 2I) 2 + I
c
1 The second equality comes from incentive compatibility, integration by parts and the fact that π(c, c) = 0 in an optimal
mechanism.
5
The next result establishes that no positive investment level is profitable for the seller, even when it
is costless. Though investment increases profits by reducing the expected cost in the case of winning the
competition, this effect is out-weighted by more disadvantageous rules imposed by the buyer, which allow
him to extract a larger fraction of the seller’s informational rent.
Proposition 7 For all I ≥ 0,
d
dI (Π(I))
< 0.
Proof. See Appendix
References
[1] Arozamena, L. and Cantillón, E. (2004), “Investment Incentives in Procurement Auctions.” Review
of Economic Studies. 71 1-18.
[2] Cisternas, G. and Figueroa, N. (2007), “Sequential Procurement Auctions and Their Effect on Investment Decisions.” Documento de Trabajo 230, CEA.
[3] Dasgupta, S.(1990) “Competition for Procurement Contracts and Underinvestment.” International
Economic Review, Vol. 31, No. 4, 841-865.
[4] Milgrom, P. (1981). “Good News and Bad News: Representation Theorems and Applications.” Bell
Journal of Economics, 380-391.
[5] Myerson, R. B. (1979). “Incentive Compatibility and the Bargaining Problem.” Econometrica. 47
61-73.
[6] Myerson, R. B. (1981). “Optimal Auction Design.” Mathematics of Operations Research. 6 58-73.
[7] Pesendorfer, M. and Jofre-Bonet, M. (2005). “Optimal Sequential Auctions.” Mimeo, LSE.
4. Appendix: Proofs
We first rewrite the procurement cost in the next lemma:
6
Lemma 8 The expected procurement cost can be written as


Ji−1 (c0 )
X Z
Y
C(I) =
fi (c)  [1 − Fl (Jl−1 (Ji (c)))] JI−1 (Ji (c))F (JI−1 (Ji (c)), I)dc
i≥2
l6=i
c

+

Y
[1 − Fl (Jl−1 (c0 ))] JI−1 (c0 )F (JI−1 (c0 ), I)
l≥2
+
Ji−1
Z (c0 )


Y
fi (c)  [1 − Fl (Jl−1 (Ji (c)))] Ji (c)[1 − F (JI−1 (Ji (c)), I)]dc
X
i≥2
l6=i
c

+c0 

Y
[1 − Fl (Jl−1 (c0 ))] [1 − F (JI−1 (c0 ), I)]
(6)
l≥2
Proof of Lemma 8: Recall that cn = (c1 , ..., cn ) and define H(cn , I) as



Y
X
X
∂F
(c1 , I)
fi (ci )
H(cn , I) ≡ JI (c1 )q1∗ (cn ) +
Jl (cl )ql∗ (cn ) + c0 1 −
qi∗ (cn )
∂c1
i≥1
l≥2
i≥2
Considering
A = {cn ∈ C n | JI (c1 ) ≤ c0 , JI (c1 ) ≤ Ji (ci ), ∀i ≥ 2}
the set of cost-vectors in which the upgrader wins the procurement auction we can write
Z
Z
n
n
C(I) = H(c , I)dc +
H(cn , I)dcn
C n \A
A
!
Set A can be written as A = A0 ∪
S
Ai
with
i≥2
A0
= {cn ∈ C n | JI (c1 ) ≤ c0 ∧ c0 < Ji (ci ), ∀i ≥ 2}
= {cn ∈ C n | c1 < JI−1 (c0 ) ∧ Ji−1 (c0 ) ≤ ci , ∀i ≥ 2}
Ai
= {cn ∈ C n | JI (c1 ) ≤ Ji (ci ) ∧ Ji (ci ) ≤ c0 ∧ (Ji (ci ) ≤ Jl (cl ), l ≥ i) ∧ (Ji (ci ) < Jl (cl ), i > l)}
= {cn ∈ C n | c1 ≤ JI−1 (Ji (ci )) ∧ ci ≤ Ji−1 (c0 ) ∧ (Jl−1 (Ji (ci )) ≤ cl , l ≥ i) ∧ (Jl−1 (Ji (ci )) < cl , i > l)}
and it is quite easy to see that Aj ∩ Ai = ∅ if i 6= j i, j ∈ {0, 2, 3, ..., n}. Note that in Ai the upgrader
wins the procurement auction and seller i reports de lowest virtual cost among all the upgrader’s rivals.
On the other hand, in A0 the same agent wins the competition but no other firm submits a bid below the
reserve cost c0 . Implicitly in our above definitions, among the lowest virtual costs, the upgrader wins the
procurement auction, which certainly doesn’t increase expected expenditures for the buyer. As a direct
consequence,
Z
X Z
H(cn , I)dcn =
H(cn , I)dcn
A
i=0,i≥2A
i
7
Now, define tl (·) ≡ Jl−1 (Ji (·)) for l ≥ 2, l 6= i and tI (·) ≡ JI−1 (Ji (·)) Integrating over Ai yields
Z
H(cn , I)dcn =
Ji−1
Z (c0 )
and observing that
Z
∂F
(c1 , I)
JI (c1 ) ∂c
1
H(cn , I)dcn =
tn (ci )
ti−1 (ci ) ti+1 (ci )
t2 (ci )
tZ
I (ci )


Y
∂F
JI (c1 )
(c1 , I)  fl (cl ) dcn
∂c1
...
...
c
Ai
Zc̄
Zc̄
Zc̄
Zc̄
= c1 +
F (c1 ,I)
∂F
∂c (c1 ,I)
1
l≥2
c
∂F
∂c1 (c1 , I)
=
d
dc1 (c1 F (c1 , I))
we obtain
Ji−1
Z (c0 )


Y
fi (c)  [1 − Fl (Jl−1 (Ji (c)))] JI−1 (Ji (c))F (JI−1 (Ji (c)), I)dc
l6=i
c
Ai
Analogously,
Z
Zc̄
H(cn , I)dcn
Zc̄
=
JI−1
Z (c0 )


Y
∂F
JI (c1 )
(c1 , I)  fl (cl ) dcn
∂c1
...
J2−1 (c0 )
A0
= 
l≥2
c
−1
Jn
(c0 )


Y
[1 − Fl (Jl−1 (c0 ))] JI−1 (c0 )F (JI−1 (c0 ), I)
(7)
l≥2
Thus,
Z
H(cn , I)dcn
=
X
i≥2
A
Ji−1
Z (c0 )

Y
fi (c)  [1 − Fl (Jl−1 (Ji (c)))] JI−1 (Ji (c))F (JI−1 (Ji (c)), I)dc
c

+

l6=i

Y
[1 − Fl (Jl−1 (c0 ))] JI−1 (c0 )F (JI−1 (c0 ), I)
(8)
l≥2
On the other hand,
C n \ An = {cn ∈ C n | (∃l ≥ 2, Jl (cl ) < JI (c1 ) ∧ Jl (cl ) ≤ c0 ) ∨ (c0 < JI (c1 ), c0 < Ji (ci ), ∀i ≥ 2)}
is the set over which the upgrader
loses the procurement auction. As before, this set can be partitioned
!
S
as C n \ A = B0 ∪
Bj with
j≥2
B0 = {cn ∈ C n | JI−1 (c0 ) < c1 ∧ Ji−1 (c0 ) < ci , ∀i ≥ 2}
(9)
Bi = {cn ∈ C n | ci ≤ Ji−1 (c0 ) ∧ (Jl−1 (Ji (ci )) ≤ cl , i ≤ l) ∧ (Jl−1 (Ji (ci )) < cl , i < l) ∧ JI−1 (Ji (ci )) < c1 }
Set B0 represents the zone in which the project is not assigned and Bi corresponds to the region where
firm i ≥ 2 wins the competition. Implicitly in the definition of these sets we assume that, in case of equal
8
lowest-virtual-costs, the task is assigned to the lowest-index competitor, which certainly doesn’t increase
expected procurement expenditures. Then we can write
Z
X Z
H(cn , I)dcn =
H(cn , I)dcn
i=0,i≥2B
i
C n \A
It is direct that
Z
H(cn , I)dcn
Ji−1
Z (c0 )
Zc̄
Zc̄
=
Zc̄
...
c
Bi
t2 (ci )
Zc̄
Zc̄
...
ti−1 (ci ) ti+1 (ci )


Y
∂F
Ji (ci )
(c1 , I)  f (cl ) dcn
∂c1
l≥2
tn (ci ) tI (ci )

Ji−1
Z (c0 )

Y
 [1 − Fl (J −1 (Ji (ci )))] [ci fi (ci ) + Fi (ci )][1 − F (J −1 (Ji (ci )), I)]dci
I
l
=
l6=i
c
Ji−1 (c0 )


Y
fi (c)  [1 − Fl (Jl−1 (Ji (c)))] Ji (c)[1 − F (JI−1 (Ji (c)), I)]dc
Z
=
(10)
l6=i
c
Also,
Z
H(cn , I)dcn
Zc̄
Zc̄
=
...
J2−1 (c0 )
B0
c0
∂F
(c1 , I) 
∂c1

Y
fl (cl ) dcn
l≥2
−1
Jn
(c0 ) JI−1 (c0 )

= c0 

Zc̄

Y
[1 − Fl (Jl−1 (c0 ))] [1 − F (JI−1 (c0 ), I)]
(11)
l≥2
As a consequence,
Z
H(cn , I)dcn
=
C n \A
X
i≥2
Ji−1
Z (c0 )

Y
fi (c)  [1 − Fl (Jl−1 (Ji (c)))] Ji (c)[1 − F (JI−1 (Ji (c)), I)]dc
c

l6=i

+c0 

Y
[1 − Fl (Jl−1 (c0 ))] [1 − F (JI−1 (c0 ), I)]
(12)
l≥2
which concludes the proof.
Proof of Proposition 6: Define


Y
αi (c) ≡ fi (c)  [1 − Fl (Jl−1 (Ji (c)))]
l6=i
9
thus, using lemma 8
C(I)
=
X
Ji−1
Z (c0 )
αi (c){JI−1 (Ji (c))F (JI−1 (Ji (c)), I) + [1 − F (JI−1 (Ji (c)), I)]Ji (c)}dc
i≥2
c

+
(13)

Y
[1 − Fl (Jl−1 (c0 ))] {JI−1 (c0 )F (JI−1 (c0 ), I) + [1 − F (JI−1 (c0 ), I)]c0 }
(14)
l≥2
Therefore, under suitable integrability conditions
C 0 (I)
2
Ji−1
Z (c0 )
∂
{F (JI−1 (Ji (c)), I)[JI−1 (Ji (c)) − Ji (c)]}dc
∂I
i≥2
c


Y
∂
{F (JI−1 (c0 ), I)[JI−1 (c0 ) − c0 ]}
+  [1 − Fl (Jl−1 (c0 ))]
∂I
X
=
αi (c)
(15)
(16)
l≥2
Define L(c, I) ≡ F (JI−1 (c), I)[JI−1 (c) − c]. Thus,
∂F −1
∂ −1
∂F −1
=
(J (c), I) (JI (c)) +
(J (c), I) [JI−1 (c) − c]
∂t I
∂I
∂I I
∂
+F (JI−1 (c), I) (JI−1 (c))
∂I
∂ −1
∂F −1
−1
−1
=
(J (c))
(J (c), I)[JI (c) − c] + F (JI (c), I)
∂I I
∂t I
∂F −1
+
(J (c), I)[JI−1 (c) − c]
∂I I
∂L
(c, I)
∂I
Recall that JI (t) = t +
F (t,I)
,
(t,I)
∂F
∂t
(17)
so, evaluating at t = JI−1 (c) we obtain
F (J −1 (c), I)
JI−1 (c) − c = − ∂F I−1
∂t (JI (c), I)
Then,
∂L
∂F −1
(c, I) =
(J (c), I)[JI−1 (c) − c]
∂I
∂I I
(18)
Therefore,
C 0 (I)
=
Ji−1
Z (c0 )
∂F −1
(J (Ji (c)), I)[JI−1 (Ji (c)) − Ji (c)]dc
∂I I
i≥2
c


Y
∂F −1
+  [1 − Fl (Jl−1 (c0 ))]
(J (c0 ), I)[JI−1 (c0 ) − c0 ]
∂I I
X
αi (c)
(19)
(20)
l≥2
−1
−1
−1
example, if we define
L(c,
I) ≡ JI (Ji (c))F (JI (Ji (c)), I) + [1 − F (JI (Ji (c)), I)]Ji (c), we need L(·, I) to be
∂L
measurable for all I and ∂I (c, I) ≤ K(c) where K ∈ L1 (C).
2 For
10
Since αi (c) ≥ 0 and JI−1 (c) − c ≤ 0, ∀c ∈ C, a sufficient condition to obtain C 0 (I) ≤ 0 is
∀i ≥ 2, ∀c ∈ [c, Ji−1 (c0 )],
and
∂F −1
(J (Ji (c)), I) ≥ 0
∂I I
∂F −1
(J (c0 ), I) ≥ 0
∂I I
which are equivalent to
∂F
(c, I) ≥ 0
∂I
since JI (c) = Ji (c) = c and JI (·) and Ji (·), i ≥ 2, are increasing functions.
∀c ∈ [c, JI−1 (c0 )],
Proof of Proposition 7: In order to show that Π(I) is strictly decreasing for all I ≥ 0, we will
d
prove that dI
(log(Π(I))) < 0, which is obviously equivalent. Recall that
2
2+I
1+I
Z2+I 1
2
c(2 + I) 2+I
(1 + I)2
Π(I) =
c dc =
1−
2
(2 + I)(3 + 2I) 2 + I
c
so, we have that log(Π(I)) satisfies
d
1
1
2
1
2+I
(log(Π(I))) =
−
−
+
log
dI
1+I
2+I
3 + 2I
(1 + I)2
2
Also, since log(x) ≤ x − 1, it is direct that log 2+I
≤ I2 . Then,
2
d
(log(Π(I))) <
dI
=
1
1
2
1
I
−
−
+
1+I
2+I
3 + 2I
(1 + I)2 2
2(1 + I)(2 + I)(3 + 2I) − 2(3 + 2I)(1 + I)2 − 4(2 + I)(1 + I)2 + I(2 + I)(3 + 2I)
2(2 + I)(3 + 2I)(1 + I)2
As a consequence it suffices to show that the numerator in the above expression is negative for any
possible I (the denominator is always strictly positive). After straightforward algebra we obtain that the
numerator is equal to −2I 3 − 5I 2 − 4I − 2, which is strictly less that zero, concluding the proof.
11